def acf_fundamental_freq(x, fmin, fmax, fs):
y = copy.copy(x)
y = preprocessing.avoid_overlap(y, N=100, f=fmax + 100, fs=fs,
plot=False) # fmax+100为的是留出一些裕度，因为低通滤波器不理想
# time_series = preprocessing.downsample(time_series, fs, 4410)
nmin = int(fs / fmax)
nmax = int(fs / fmin)
acf = stattools.acf(y, nlags=nmax)
f0 = fs / np.argmax(acf[nmin:])
return f0
sample_num = [] #统计每个标签的样本数
for j in range(len(labelname)):
subpath = path + '\\' + labelname[j]
subfilename = os.listdir(subpath) #查看音频文件目录
sample_num.append(len(subfilename)) # 统计文件数量，也可以认为是统计样本数量
for i in range(len(subfilename)):
audio_data, sample_rate = librosa.load(subpath + '\\' + subfilename[i],
sr=None,
mono=True,
res_type='kaiser_best') # 读取文件
# pre_amphasis = preprocessing.pre_emphasis(audio_data, 0.97, pic=savepic + '\\' + 'pre_amphasis_'+str(j)+'_'+str(i))
pre_amphasis = audio_data
avoid_overlap = preprocessing.avoid_overlap(pre_amphasis,
N=10,
f=4000,
fs=sample_rate,
plot=False)
downsample = preprocessing.downsample(avoid_overlap, sample_rate,
downsample_rate)
silence_remove = preprocessing.silence_remove(
downsample,
limit=np.max(downsample) / 20,
option='hilbert',
# pic=savepic + '\\' + 'silence_remove_hilbert_' + str(j)+'_'+str(i)
pic=None)
# _, _, _ = visual.stft_specgram(silence_remove, picname=savepic + '\\' + 'stft_'+str(j)+'_'+str(i), fs=downsample_rate,
# nperseg=512, noverlap=128, nfft=1024)
buffer = [j] + list(silence_remove)
# 写入数据
import scipy.signal as signal
import numpy as np
import librosa as lib
import matplotlib.pyplot as plt
import time
from statsmodels.tsa import stattools
import preprocessing
import feature_extraction as fe
# ex = '..\\..\\boy_and_girl\\class1\\arctic_a0012.wav'
# ex = '..\\..\\cello_and_viola\\viola\\Viola.arco.ff.sulA.A4.stereo.aiff'
time_series, fs = lib.load(ex, sr=None, mono=True, res_type='kaiser_best')
time_series = preprocessing.avoid_overlap(time_series,
N=100,
f=500,
fs=fs,
plot=False)
time_series = preprocessing.downsample(time_series, fs, 4410)
print(fs)
frames = preprocessing.frame(time_series, int(0.03 * fs), int(0.015 * fs))
for i in range(frames.shape[1]):
acf1 = stattools.acf(frames[:, i], nlags=100)
fft, _ = fe.fft_singleside(frames[:, i], 4410, 8096)
plt.figure()
plt.subplot(211)
plt.plot(np.abs(fft))
plt.subplot(212)
plt.stem(acf1)
import librosa as lib
import matplotlib.pyplot as plt
import time
from preprocess_filter import *
import preprocessing
import feature_extraction as fe
# ex = '..\\..\\数据集2\\pre2012\\bflute\\BassFlute.ff.C5B5.aiff'
ex = '..\\..\\cello_and_viola\\viola\\Viola.arco.ff.sulA.A4.stereo.aiff'
x, fs = lib.load(ex, sr=None, mono=True, res_type='kaiser_best')
x = x[1000:2500]
downsample_rate = 22050
avoid_overlap = preprocessing.avoid_overlap(x,
N=20,
f=downsample_rate / 2,
fs=fs,
plot=False)
downsample = preprocessing.downsample(avoid_overlap, fs, downsample_rate)
S1, f1 = fe.fft_singleside(x, fs, n=8192)
S2, f2 = fe.fft_singleside(avoid_overlap, fs, n=8192)
S3, f3 = fe.fft_singleside(downsample, downsample_rate, n=8192)
plt.figure()
ax1 = plt.subplot(211)
ax1.set_xlabel('time')
ax1.set_ylabel('mag')
ax1.set_title('original signal')
plt.plot(x)
ax2 = plt.subplot(212)
for i in range(1):
subpath = path + '\\' + labelname[i]
subfilename = os.listdir(subpath)
sample_num.append(len(subfilename)) # 统计文件数量，也可以认为是统计样本数量
list = []
audio_data, sample_rate = librosa.load(
subpath + '\\' + subfilename[0], sr=None, mono=True, res_type='kaiser_best')
list.append(audio_data.tolist())
audio_data_set[labelname[i]] = list
for i in range(1):
data = np.array(audio_data_set[labelname[i]][0])
pre_amphasis = preprocessing.pre_emphasis(data, 0.97, pic='pre_amphasis'+str(i))
avoid_overlap = preprocessing.avoid_overlap(pre_amphasis,
N=10,
f=11000,
fs=sample_rate,
plot=True)
downsample = preprocessing.downsample(avoid_overlap, sample_rate, downsample_rate)
start = time.perf_counter()
silence_remove = preprocessing.silence_remove(
downsample,
limit=0.001,
option=filter,
pic='silence_remove_filter'+str(i),
N=10,
f=100,
fs=downsample_rate,
plot=True)
end = time.perf_counter()
print(end-start)
def load_and_preprocess(amphasis, down, clip, factor, saveprojectpath,
savedata, savepic, savetestdata, savepreprocess,
savefeature, path, downsample_rate, frame_time,
frame_length, frame_overlap, test_rate):
if not os.path.exists('..\\仿真结果'):
os.mkdir('..\\仿真结果')
if not os.path.exists(saveprojectpath):
os.mkdir(saveprojectpath)
if not os.path.exists(savedata): # 保存数据
os.mkdir(savedata)
if not os.path.exists(savepic): # 创建保存图片的文件夹
os.mkdir(savepic)
if not os.path.exists(savetestdata): # 创建保存测试数据的文件
os.mkdir(savetestdata)
# 写入数据
with open(savepreprocess, 'w',
encoding='utf-8') as csvfile: # 先创建csv文件，什么都不写入，之后再追加
writer = csv.writer(csvfile)
# 读取音频文件目录
labelname = os.listdir(path) # 获取该路径下的子文件名，也是标签的名字
for j in range(len(labelname)):
subpath = path + '\\' + labelname[j] # 数据集的子文件路径
subfilename = os.listdir(subpath) # 查看音频文件目录
for i in range(len(subfilename)):
audio_data, sample_rate = librosa.load(
subpath + '\\' + subfilename[i],
sr=None,
mono=True,
res_type='kaiser_best') # 读取音频文件
audio_data = librosa.util.normalize(audio_data,
norm=np.inf,
axis=0,
threshold=None,
fill=None)
# audio_data = audio_data[0:8820]
##########################################################################################################################
if amphasis:
pre_amphasis = preprocessing.pre_emphasis(audio_data,
0.97,
pic=None)
# pic=savepic + '\\' + 'pre_amphasis_'+str(j)+'_'+str(i))
else:
pre_amphasis = audio_data
if down:
avoid_overlap = preprocessing.avoid_overlap(pre_amphasis,
N=20,
f=downsample_rate /
2,
fs=sample_rate,
plot=False)
downsample = preprocessing.downsample(avoid_overlap,
sample_rate,
downsample_rate)
else:
downsample = pre_amphasis
if clip == 'hilbert':
silence_remove = preprocessing.silence_remove(
x=downsample,
limit=np.max(downsample) * factor,
fs=downsample_rate,
option='hilbert',
# pic=savepic + '\\' + 'silence_remove_hilbert_' + str(j)+'_'+str(i))
pic=None)
elif clip == 'HF':
silence_remove = preprocessing.silence_remove(
x=downsample,
limit=np.max(downsample) * factor,
fs=downsample_rate,
option='HF',
# pic=savepic + '\\' + 'silence_remove_hilbert_filter_' + str(j)+'_'+str(i))
pic=None)
elif clip == 'filter':
silence_remove = preprocessing.silence_remove(
x=downsample,
limit=0.02,
option='filter',
pic=savepic + '\\' + 'silence_remove_filter_' + str(j) +
'_' + str(i),
N=10,
f=600,
fs=downsample_rate)
else:
silence_remove = downsample
########################################################################################################################
# j表示标签， i表示同一标签下的音频文件序号
buffer = [j] + [i] + list(silence_remove)
with open(savepreprocess, 'a+', newline='',
encoding='utf-8') as csvfile:
csv_write = csv.writer(csvfile)
csv_write.writerow(buffer) # 逐行写入数据
print('preprocessing:', j, i)
return labelname
